Ahmad Said Email and Phone Number

• Developed a TypeScript web application with React and Node.js, integrating LLM agents for automated epidemiology analysis. Achieved a 7x increase in processing speed and significant cost savings in real-time epidemic tracking.• Was lead engineer of a team of 6 in implementing an in-house alternative to Chat-GPT, hosted on EC2 with ELB. Launched our beta and leading efforts to reduce costs and increase user engagement.• Leveraged PostgreSQL database and Cloudwatch to log chat history and user metrics.

• Prototyped multiple LLMs to design a proof of concept for Expedia Chat leveraging LLMs to help customers with broad coverage of customer services and shopping experience intents.• Enhanced understanding capabilities and handled all experimentation, data collection, fine-tuning, and design.• Gathered 5,000 lines of data to fine-tune using SQL queries written on Tableau• Provided a production-ready LLM, hosted on Sagemaker and deployed on integration environment, that is grounded with real-time information from API's.

• Developed and improved on a novel method to eliminate AI Bias in tabular datasets which outperforms all current methods and is published in SPIE. • Designed and ran experiments on large datasets to test the efficacy of the method and other competing methods. Completely eliminated bias in 11/12 datasets by reducing all metrics by around 90%. • Shipped an automated script that can handle any dataset passed, and output metric comparisons between the improvements and original.

• Prototyped various neural networks and reconciled LIDAR and image data to increase accuracy of a self-driving vehicle. • Detected and tracked vehicles, pedestrians, and bicyclists on over 200gb of image and video data for autonomous cars. Achieved an accuracy of 70% and precision of 89%.

• Used Faster-RCNN and YOLOv6 to identify and crop QR codes in 3 TB of underwater images, and applied edge detection, Hough transform, and deblurGAN along with various image processing and deep learning techniques to increase decoding accuracy from 3% to 85% of all readable labels.• Returned the location of QR code containers underwater with a degree of accuracy of 97% and a precision of 99%.